Rewind

Inspiration

At some point or another, everyone has the desire to relive their most precious memories. Whether it is a late-night hackathon grind with friends or an overseas trip, we constantly seek new ways to surround ourselves with nostalgia. To fit the theme of this year's hackathon, we decided to build Rewind, a device that verbally narrates your memories in the wording and voice of David Attenborough.

What it does

Memory Lane is a hardware boombox that can take in images and personal journals and turn them into the script of a David Attenborough documentary. Then, using voice cloning, it narrates your memories in an old-fashioned radio station vibe.

How we built it

We utilize a pipeline of AI models starting with GPTVision to generate image descriptions. We then took those descriptions and used prompt engineering to generate a monologue in the voice of David Attenborough using Cohere's Generate API. Finally, we use Eleven Labs to create super realistic celebrity voices. The hardware was built on an Arduino Uno with speakers and an amplifier.

Challenges we ran into

Choosing the right prompts and finding the best models were crucial to creating meaningful nostalgic stories, and to hear your life from another point of view.

Many challenges arose during the hardware component. Finding libraries to generate a voice out of $1 speakers was a pain. We swapped through multiple boards, and circuit configurations, feeling like hamsters going in circles at times. Setting up our amplifier with zero documentation and diagrams was a mental stretch, especially considering this was our team's first hardware hack. Finally, attempting to surpass the memory limits of Arduino boards proved most difficult. We had to do boolean debugging across serial ports and test the limits for sending data to an Arduino externally.

Accomplishments that we're proud of

This is the first hardware project any member of our team has submitted. We decided to take the opportunity of UofTHacks to explore a new dimension of hackathons beyond just web applications and games. While the parts sourcing and debugging were incredibly time-consuming and difficult we are proud of the semi-polished device we produced.

What we learned

We learned that AI tools combined in the right ways can bring less of the "oh, that's cool for an AI" reaction, and move towards a meaningful connection to life. Also, hardware is always a pain. It's terribly documented. And you will never have enough components for your project. But pushing yourself to learn new technologies leads to the most growth.

What's next for Boombox Narrator

We plan to add a camera module/extension to the boombox. Since the form factor of the device is comparable to a handheld camera, an integrated webcam would allow users to capture and convert their moments within one device. We also want to build a smartphone integration to allow an airdrop-like system of transferring photos and digital journals directly to the boombox.